Download Relationship among phenotypic plasticity

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Adaptive evolution in the human genome wikipedia , lookup

Site-specific recombinase technology wikipedia , lookup

Pharmacogenomics wikipedia , lookup

Point mutation wikipedia , lookup

Public health genomics wikipedia , lookup

Artificial gene synthesis wikipedia , lookup

Genetic drift wikipedia , lookup

Genome evolution wikipedia , lookup

Gene expression profiling wikipedia , lookup

Genetic engineering wikipedia , lookup

Behavioural genetics wikipedia , lookup

Species distribution wikipedia , lookup

Dual inheritance theory wikipedia , lookup

Designer baby wikipedia , lookup

Human genetic variation wikipedia , lookup

Nutriepigenomics wikipedia , lookup

Heritability of IQ wikipedia , lookup

Genome (book) wikipedia , lookup

Epistasis wikipedia , lookup

Gene expression programming wikipedia , lookup

Population genetics wikipedia , lookup

Quantitative trait locus wikipedia , lookup

Koinophilia wikipedia , lookup

Microevolution wikipedia , lookup

Transcript
Relationship among phenotypic plasticity, phenotypic fluctuations, robustness, and evolvability
529
Relationship among phenotypic plasticity, phenotypic fluctuations,
robustness, and evolvability; Waddington’s legacy revisited under the
spirit of Einstein
KUNIHIKO KANEKO
Department of Pure and Applied Sciences, University of Tokyo, 3-8-1 Komaba, Meguro-ku, Tokyo 153-8902, Japan
and
ERATO Complex Systems Biology Project, JST, 3-8-1 Komaba, Meguro-ku, Tokyo 153-8902, Japan
(Email, [email protected])
Questions on possible relationship between phenotypic plasticity and evolvability, and that between robustness and
evolution have been addressed over decades in the field of evolution-development. Based on laboratory evolution
experiments and numerical simulations of gene expression dynamics model with an evolving transcription network,
we propose quantitative relationships on plasticity, phenotypic fluctuations, and evolvability. By introducing an
evolutionary stability assumption on the distribution of phenotype and genotype, the proportionality among phenotypic
plasticity against environmental change, variances of phenotype fluctuations of genetic and developmental origins, and
evolution speed is obtained. The correlation between developmental robustness to noise and evolutionary robustness
to mutation is analysed by simulations of the gene network model. These results provide quantitative formulation on
canalization and genetic assimilation, in terms of fluctuations of gene expression levels.
[Kaneko K 2009 Relationship among phenotypic plasticity, phenotypic fluctuations, robustness, and evolvability; Waddington’s legacy revisited
under the spirit of Einstein; J. Biosci. 34 529–542]
1.
1.1
Introduction
Questions on evolution
(i) How is evolvability characterized? Is it related with
developmental process? Naively speaking, one may expect
that if developmental process is rigid, phenotype is not
easily changed, so that evolution does not progress so fast.
Then one may expect that evolvability is tightly correlated
with phenotypic plasticity. Here the plasticity refers to the
changeability of phenotype as a result of developmental
dynamics (Callahan et al. 1997; West-Eberhard 2003;
Kirschner and Gerhart 2005; Pigliucci et al. 2006). If there
is correlation between plasticity and evolvability (Ancel and
Fontana 2002), one may formulate relationship between
evolvability and development quantitatively.
(ii) Besides plasticity, an important issue in evolution is
robustness. Robustness is defined as the ability to function
against possible changes in the system (Barkai and Leibler
1997; Wagner et al. 1997; Alon et al. 2000; Wagner 2000,
2005; Ciliberti et al. 2007; Kaneko 2007). In any biological
system, these changes have two distinct origins: genetic
and epigenetic. The former concerns structural robustness
of the phenotype, i.e. rigidity of phenotype against genetic
changes produced by mutations. On the other hand, the latter
concerns robustness against the stochasticity that can arise in
environment or during developmental process, which includes
fluctuation in initial states and stochasticity occurring during
developmental dynamics or in the external environment.
Now, there are several questions associated with
robustness and evolution. Is robustness increased or
decreased through evolution? In fact, Schmalhausen (1949)
proposed stabilizing evolution and Waddington (1942,
1957) proposed canalization. Both of them discussed the
possibility that robustness increases through evolution. On
the other hand, since the robustness increases rigidity in
phenotype, the plasticity as well as the evolvability may be
decreased accordingly through evolution.
Besides these robustness-evolution questions, another
issue to be addressed is the relationship between the two
Keywords. Evolvability; genetic assimilation; phenotypic fluctuation; phenotypic plasticity
J. Biosci. 34(4), October 2009
J. Biosci. 34(4), October 2009, 529–542, © Indian Academy of Sciences 529
Kunihiko Kaneko
530
types of robustness, genetic and epigenetic. Do the two types
of robustness increase or decrease in correlation through
evolution? In fact, the epigenetic robustness concerns with
developmental stability. Higher epigenetic robustness is,
more stable the phenotype in concern shaped by development
is. On the other hand, genetic robustness concerns with
the stability against mutational change, i.e. evolutionary
stability. Hence the question on a possible relationship
between genetic and epigenetic robustness concerns with
the issue on evolution-development relationship, that is an
important topic in biology.
1.2
Formulation of the problem in terms of
stochastic dynamical systems
Here we study the questions raised above, in terms of
fluctuations. Let us first discuss epigenetic robustness,
i.e. the degree of phenotype change as a result of
developmental process. The developmental dynamics to
shape a phenotype generally involves stochasticity. As has
been recognized extensively, gene expression dynamics are
noisy due to molecular fluctuation in chemical reactions.
Accordingly, the phenotype, as well as the fitness, of
isogenic individuals is distributed (Oosawa 1975; Spudich
and Koshland 1976). Still, too large variability in the
phenotype relevant to fitness should be harmful to the
survival of the organisms. Phenotype that is concerned
with fitness is expected to keep some robustness against
such stochasticity in gene expression, i.e. robustness in
‘developmental’ dynamics to noise.
When the phenotype is less robust to noise, this
distribution is broader. Hence, the variance of isogenic
phenotypic distribution, denoted as Vip here, gives an
index for robustness to noise in developmental dynamics
(Kaneko 2006, 2007). For example, distributions of protein
abundances over isogenic individual cells have been
measured, by using fluorescent proteins. The fluorescence
level which gives an estimate of the protein concentration
is measured either by flow cytometry or at a single cell
microscopy (McAdams and Arkin 1997; Hasty et al. 2000;
Elowitz et al. 2002; Bar-Even et al. 2004; Furusawa et al.
2005; Kaern et al.2005; Krishna et al. 2005).
Genetic robustness is also measured in terms of
fluctuation. Due to mutation to genes, the phenotype
(fitness) is distributed. Since even phenotype of isogenic
individuals is distributed, the variance of phenotype
distribution of heterogenic population includes both the
contribution from phenotypic fluctuation in isogenic
individuals and that due to genetic variation. To distinguish
the two, we first obtain the average phenotype over isogenic
individuals, and then compute the variance of this average
phenotype over heterogenic population. Then this variance
J. Biosci. 34(4), October 2009
is due only to genetic heterogeneity. This variance is
denoted as Vg here. Then, the robustness to mutation is
estimated by this variance. If Vg is smaller, genetic change
gives little influence on the phenotype, and larger genetic (or
mutational) robustness is implied.
Phenotypic plasticity, on the other hand, represents
the degree of change in phenotype against variation
in environment. Quantitatively, one can introduce the
following “response ratio” as a measure of plasticity.
It is defined by the ratio of change in phenotype to
environmental change. For example, measure the (average)
concentration (x) of a specific protein. We measure this
concentration by (slightly) changing the concentration
(s) of an external signal molecule. Then, response ratio
R is computed as the change in the protein concentration
divided by the change in the signal concentration (R = dx/
ds)1. This gives a quantitative measure for the phenotypic
plasticity. In general, there can be a variety of ways to
change the environmental condition. In this case, one may
also use the variance of such response ratios over a variety
of environmental changes, as a measure of phenotypic
plasticity.
By using these variances Vip, Vg and plasticity the questions
we address are represented as follows (Gibson and Wagner
2000; de Visser et al. 2000; Weinig 2000; Ancel and Fontana
2002; West-Eberhard 2003; Kirschner and Gerhart 2005):
(i) Does the evolution speed increase or decrease
with robustness?— relationship between evolution
speed with Vg or Vip.
(ii) Are epigenetic and genetic robustness related?
— relationship between Vg and Vip.
(iii) Does robustness in-(de-)crease through evolution?
— change of Vg or Vip through evolution.
(iv) How is phenotypic plasticity related with
robustness and evolution speed? — relationship
between R and Vg or Vip.
To answer these questions, we have carried out evolution
experiments by using bacteria (Sato et al. 2003) and
numerical simulations on catalytic reaction (Kaneko and
Furusawa 2006) and gene networks (Kaneko 2007, 2008),
while we have also developed a phenomenological theory
on the phenotype distribution. Here we briefly review
these results and discuss their relevance to evolutiondevelopment.
1
As will be discussed in Section 2.3, it is often relevant to define
the phenotype variable after taking a logarithm. In this case a phenotype variable is defined as x = log(X) with X as the original variable, say, the protein concentration. Likewise, external signal often
works in a logarithmic scale, so that the external parameter s is also
defined as log(S), with S as a signal concentration. In this case R =
dlog(X)/dlog(S), which, indeed, is often adopted in cell biology.
Relationship among phenotypic plasticity, phenotypic fluctuations, robustness, and evolvability
2. Macroscopic approach I:
Fluctuation-response relationship
One might suspect that isogenic phenotype fluctuations
Vip are not related to evolution, since phenotypic change
without genetic change is not transferred to the next
generation. However, the variance, a degree of fluctuation
itself, can be determined by the gene, and accordingly it is
heritable (Ito et al. 2009). Hence, there may be a relationship
between isogenic phenotypic fluctuation and evolution. In
fact, we have found evolutionary fluctuation–response
relationship in bacterial evolution in the laboratory, where
the evolution speed is positively correlated with the variance
of the isogenic phenotypic fluctuation (Sato et al. 2003).
This correlation or proportionality was confirmed in a
simulation of the reaction network model (Kaneko and
Furusawa 2006). The origin of proportionality between
the isogenic phenotypic fluctuation and genetic evolution
has been discussed in light of the fluctuation–response
relationship in statistical physics (Einstein 1926; Kubo
et al. 1985). In the present section we briefly summarize
these recent studies.
2.1
Selection experiment by bacteria
We carried out a selection experiment to increase the
fluorescence in bacteria and checked a possible relationship
between evolution speed and isogenic phenotypic
fluctuation. First, by attaching a random sequence to the N
terminus of a wild-type Green Fluorescent Protein (GFP)
gene, protein with low fluorescence was generated. The
gene for this protein was introduced into Escherichia coli,
as the initial generation for the evolution. By applying
random mutagenesis only to the attached fragment in the
gene, a mutant pool with a diversity of cells was prepared.
Then, cells with highest fluorescence intensity are selected
for the next generation. With this procedure, the (average)
fluorescence level of selected cells increases by generations.
The evolution speed at each generation is computed as
the difference of the fluorescence levels between the two
generations. Then, to observe the isogenic phenotypic
fluctuation, we use a distribution of clone cells of the selected
bacteria. The distribution of the fluorescence intensity of the
clones was measured with the help of flow cytometry, from
which the variance of fluorescence was measured.
From these data we plotted the evolution speed divided
by the synonymous mutation rate versus the fluorescence
variance2. The data support strong correlation between the
two, suggesting proportionality between the two (Sato et al.
2003).
2
To be precise, the phenotype variable x is defined by log (fluorescence), as will be discussed in Section 2.3.
J. Biosci. 34(4), October 2009
2.2
531
Supports from numerical experiments
To confirm this relationship between the evolution speed and
isogenic phenotypic fluctuation quantitatively, we have also
numerically studied a model of reproducing cells consisting
of catalytic reaction networks. Here the reaction networks of
mutant cells were slightly altered from the network of their
mother cells. Among the mutants, those networks with a
higher concentration of a given, specific chemical component
were selected for the next generation. Again, the evolution
speed was computed by the increase of the concentration
at each generation, and the fluctuation by the variance of
the concentration over identical networks3. From extensive
numerical experiments, the proportionality between the two
was clearly confirmed (Kaneko and Furusawa 2006).
2.3
Remark on Gaussian type distribution
In the above experiment and numerical model, the
fluorescence or the concentration obeys approximately
the log-normal distribution, i.e. log(fluorescence) or
log(concentration) obeys nearly the Gaussian distribution. In
the following analysis that we borrow from statistical physics,
the distribution is assumed to be nearly Gaussian. If the
distribution is of the log-normal type, we adopt such quantity
taking a logarithm, so that the distribution of the phenotypic
variable is nearly Gaussian, in order to apply the theoretical
discussion. In fact, Haldane suggested that most phenotypic
variables should be defined after taking a logarithm. In the
above analyses between fluctuation and response, all the
quantities are defined and computed after taking a logarithm
log(fluorescence) and log(concentration) are used to compute
the evolution speed and the variance. The linear relationship
between the two is confirmed by using these variables.
2.4
Fluctuation-response relationship
To understand the proportionality between the phenotypic
fluctuation and the evolution speed, we first discuss a
possible general relationship between fluctuation and
response. Here, we refer to a measurable phenotype quantity
(e.g. logarithm of the concentration of a protein) as a
“variable” x of the system, while we assume the existence
of a “parameter” which controls the change of the variable.
The parameter corresponding to the variable x is indicated
as a. In the case of adaptation, this parameter is a quantity
to specify the environmental condition, while, in the case
of evolution, the parameter is an index of genotype that
governs the corresponding phenotypic variable.
3
Again, the phenotypic variable x is defined by log(concentration),
and the variance is measured by the distribution of log
(concentration) (see Section 2.3).
Kunihiko Kaneko
532
Even among individuals sharing the identical
gene (the parameter) a, the variables x are distributed.
The distribution P(x;a) of the variable x over cells
is defined, for a given parameter a (e.g. genotype).
Here the “average” and “variance” of x are defined
with regards to the distribution P(x;a) (see figure 1
for schematic representation of the fluctuation-response
relationship).
Consider the change in the parameter value from a to a→
a+δa. Then, the proposed fluctuation-response relationship
(Sato et al. 2003; Kaneko 2006) is give by
< x >a +Δa − < x >a
∝< (δx )2 >,
Δa
(1)
where <x>a and <(δx)2> = <(x–<x>)2> are the average and
variance of the variable x for a given parameter value a,
respectively.
The above relationship is derived by assuming that the
distribution P(x;a) is approximately Gaussian and that the
effect of the change in a on the distribution is represented by
a bi-linear coupling between x and a. With this assumption,
the distribution is written by
P (x ; a ) = N 0exp(−
(x − X 0 )2
2α(a )
+ v(x , a )),
(2)
with N0 a normalization constant so that ∫P(x : a)dx = 1.
Here, X0 is the peak value of the variable at a = a0, where
the term v(x, a) gives a deviation from the distribution at
a = a0, so that v(x, a) can be expanded as V(x,a) =
C(a–a0)(x–X0)+…, with C as a constant, where … is a higher
order term in (a–a0) and (x–X0) which will be neglected in
the following analysis.
Assuming this distribution form, the change of the
average value <x> following the change of the parameter
from a0 to a0 + Δa is straightforwardly computed, which
leads to
< x >a =a
0
+Δa
− < x >a =a0
Δa
= C α(a 0 + Δa ).
(3)
Noting α = <(δx)2>, and neglecting deviation between
α(a0 + Δa) and α(a0), we get
< x >a =a
0
+Δa
− < x >a =a0
Δa
= C < (δx )2 > .
(4)
This relationship is a general result that holds for Gaussianlike distributions if changes in parameters are represented
by a “linear coupling term”, which brings about a shift of the
average of the corresponding variable.
J. Biosci. 34(4), October 2009
Remark: Although the above formula is formally
similar to that used in statistical physics, it is not
grounded on any established theory. It is a proposal
here. We set up a phenomenological description so
that the formula is consistent with the data obtained by
experiments and numerical simulations on reaction network
models. (However, such phenomenological description
is the first necessary step in science, as the history of
thermodynamics tells.) The choice of the distribution form
is not derived from the first principle, but is based on several
assumptions.
First, the distribution must be close to Gaussian, or the
variable has to be scaled properly so that the corresponding
distribution is nearly Gaussian. If the variance of
the distribution of the variable in concern is not very
large, by suitable choice of variable transformation,
the distribution can be transformed to be nearly Gaussian.
As mentioned in Section 2.3, log-normal distribution is
often observed in phenotypic variables z in a biological
system, say the concentration of some protein. In this case,
one can just adopt x = log(z) as a phenotypic variable in
concern.
Second, the choice of the coupling form in eq. (2) is also
an assumption so that the fluctuation-response relation is
consistent with experimental and numerical observations.
With this bi-linear coupling form C(x–X0)(a – a0), the
relationship is obtained. In this sense, eq. (4) is not derived,
but we choose a phenomenological distribution function (2)
so that eq. (4) is reached to be consistent with the observations
in experiments or numerical models. Here, neglect of
higher-order terms as (x–X0)k (a–a0)m (with k, m > 1)
will be justified if the deviations from a0 and X0 are small,
but the validity of such expansion in the distribution itself
is a proposal here. In other words, possible dependence of
C on the parameter a is discarded, so that C is regarded as a
constant. This will be justified if the change in a in concern
is not so large.
2.5
Plasticity-fluctuation relationship
First we apply the above relation to plasticity. Here we make
the following correspondence:
(i)
(ii)
(iii)
(iv)
(v)
a — parameter to characterize the environmental
condition
change in a — change in environment
<x> — a variable characterizing a phenotypic state
change in <x> — change in the average phenotype
due to the environmental change
<(δx)2> — variance of the phenotypic distribution
at a given fixed environmental condition.
Then, the relationship eq. (1) means that the plasticity is
proportional to the fluctuation.
Relationship among phenotypic plasticity, phenotypic fluctuations, robustness, and evolvability
Figure 1. Schematic diagram on fluctuation-response
relationship. The distribution function P(x;a) is shifted by the
change of the parameter a. If the variance is larger the shift is
larger.
2.6
Evolutionary fluctuation-response relationship
This fluctuation-response relationship can be applied to
evolution by making the following correspondence:
(i) a — a parameter that specifies the genotype
(ii) change in a — genetic mutation
(iii) change in <x> — change in the average phenotype
due to the genetic change, and
(iv) <(δx)2> — variance of the phenotypic distribution
over clones Vip.
Then, the above relationship (1) suggests that the evolution
speed of a phenotype (i.e. the change of the average
phenotype per generation) divided by the mutation rate
is proportional to Vip, the variance of isogenic phenotypic
fluctuation. This is what we have obtained in the experiment
and simulations.
The above proposition can be interpreted as follows: As
the degree of the fluctuations of the phenotype decreases, the
change of the phenotype under a given rate of genetic change
also decreases. In other words, the larger the phenotypic
fluctuation, the larger the evolution speed is. In this sense,
Vip can work as a measure for the degree of evolvability.
3.
Macroscopic approach II: Stability theory for
genotype-phenotype mapping
533
and phenotypic fluctuation. It is the so-called fundamental
theorem of natural selection by Fisher (1930) which states
that evolution speed is proportional to Vg, the variance of the
phenotypic fluctuation due to genetic variation. It is given
by the variance of the average phenotype for each genotype
over a heterogenic population. In contrast, the evolutionary
fluctuation – response relationship, proposed above,
concerns phenotypic fluctuation of isogenic individuals as
denoted by Vip. While Vip is defined as a variance over clones,
i.e., individuals with the same genes, Vg is a result of the
heterogenic population distribution. Hence, the fluctuation–
response relationship and the relationship concerning Vg by
Fisher’s theorem are not identical.
If Vip and Vg are proportional through an evolutionary
course, the two relationships are consistent. Such
proportionality, however, is not self-evident, as Vip is related
to variation against the developmental noise and Vg is against
the mutation. The relationship between the two, if it exists,
postulates a constraint on genotype–phenotype mapping
and may create a quantitative formulation of a relationship
between development and evolution.
Here we formulate a phenomenological theory so that
the evolutionary fluctuation-response relationship and
Fisher’s theorem is consistent (see figure 2). Recall that
originally the phenotype is given as a function of gene, so
that we considered the conditional distribution function
P(x;a). However, the genotype distribution is influenced
by phenotype through the selection process, as the fitness
for selection is a function of phenotype. Now, consider a
gradual evolutionary process. Then, it could be possible to
assume that some “quasi-static” distribution on genotype and
phenotype is obtained as a result of feedback process from
phenotype to genotype (see figure 2). Considering this point,
we hypothesize that there is a two-variable distribution P(x,
a) both for the phenotype x and genotype a.
By using this distribution, Vip, variance of x of the
distribution for given a, can be written as Vip (a) = ∫(x–x(a))2
P(x, a)dx, where x (a) is the average phenotype of a clonal
population sharing the genotype a, namely x(a) = ∫P(x,
a)xdx. Vg is defined as the variance of the average x (a),
over genetically heterogeneous individuals and is given
by Vg = ∫(x(a)–<x– >)2 p(a)da, where p(a) is the distribution
of genotype a and <x– > as the average of x (a) over all
genotypes.
Assuming the Gaussian distribution again, the distribution
P(x, a) is written as follows:
exp[−
P (x , a ) = N
(x − X 0 )2
2α(a )
+
(5)
1
(a − a 0 )2 ],
2μ
with N̂ as a normalization constant. The Gaussian distribution exp(− 21μ (a − a 0 )2 ) represents the distribution of genoC (x − X 0 )(a − a 0 )) −
The evolutionary fluctuation-response relationship mentioned above casts another mystery to be solved. There
is an established relationship between evolutionary speed
J. Biosci. 34(4), October 2009
Kunihiko Kaneko
534
types at around a = a0, whose variance is (in a suitable
unit) the mutation rate μ. The coupling term C(x–X0) (a–a0)
represents the change of the phenotype x by the change of
genotype a. Recalling that the above distribution (5) can be
rewritten as
exp[−
P (x , a ) = N
(x − X 0 − C (a − a 0 )α(a ))2
2α(a )
C α(a )
1
+(
− )(a − a 0 )2 ],
2
2μ
2
(6)
the average phenotype value for given genotype a
satisfies
xa ≡
∫ xP(x, a )dx = X
0
+ C (a − a 0 )α(a ).
(7)
Now we postulate evolutionary stability, i.e. at each stage of
the evolutionary course, the distribution has a single peak in
(x, a) space. This assumption is rather natural for evolution
to progress robustly and gradually. When a certain region of
phenotype, say x > xthr, is selected, the evolution to increase
x works if the distribution is concentrated. On the other
hand, if the peak is lost and the distribution is flattened, then
selection to choose a fraction of population to increase x no
longer works, as the distribution is extended to very lowfitness values (see also a remark at the end of this section).
In the above form, this stability condition is given by
1
αC 2
−
≤ 0, i.e.
2
2μ
μ≤
1
≡ μmax .
αC 2
(8)
This means that the mutation rate has an upper bound
μ max beyond which the distribution does not have a peak
in the genotype–phenotype space. In the above form, the
distribution becomes flat at μ = μ max, so that mutants with
low fitness rate appear, as discussed as error catastrophe
by Eigen (Eigen and Schuster 1979) (see also the remark
at the end of this section). Recalling Vg = <(x(a)–X0)2> and
eq.(6), Vg is given by (Cα)2 < (δa)2>. Here, (δa)2 is computed
by the average over P(x,a), so that it is obtained by
N
∫ (a − a )
2
0
exp[( 2αC(a ) − 21μ )(a − a 0 )2 ]da .
2
From this formula we obtain
Vg =
μ
μC 2α2
μmax
=
α
.
μ
1 − μC 2α
1 − μmax
(9)
If the mutation rate μ is small enough to satisfy μ« μ max, we
get
Vg ∼
μ
V ,
μmax ip
J. Biosci. 34(4), October 2009
(10)
by recalling that Vip (a) = α(a). Thus we get the proportionality
between Vip and Vg.
Note that if we choose the distribution consisting of
heterogenic individuals taking a given phenotype value
x only, the variance <(a–a0)2>fixed–x is given simply by μ.
The variance of average phenotype over such population
sharing the phenotype value x is defined as Vig (Kaneko and
Furusawa 2008). Then, by using eq. (7) we get Vig = (αC)2 <
(a–a0)2 >fixed–x = μ (αC)2, instead of eq.(9). Accordingly, the
inequality (8) is rewritten as
Vip ≥ Vig,
(11)
μ
μmaxVig
while Vip =
is also obtained without assuming μ «
μ max. If the evolution process successively increases the
selected value of x by generations, the distribution at each
generation is centered at around a given phenotype. In such
case, Vig, instead of Vg, may work as a relevant measure for
the phenotypic variance due to genetic change. Then the
inequality (11) can work as a measure for stability.
Note again that the above relationships (10) and (11) are
not derived from the first principle, but a phenomenological
description to be consistent with numerical experiments.
There are several assumptions to obtain the relationships.
First, whether genotype is represented by a continuous
parameter a is not a trivial question, as gene is coded
by a genetic sequence. Possible change by mutation to
the sequence can be represented by a walk along a highdimensional hyper-cubic lattice, and whether one can
introduce a scalar parameter a instead is not self-evident.
A candidate for such scalar parameter is Hamming distance
from the fittest sequence, or a projection according to the
corresponding phenotype (Sato and Kaneko 2007). We
note that such scalar genetic parameter is often adopted in
population genetics also.
Second, description by using a two-variable distribution
in genotype and phenotype P(x,a) is not trivial, either. As
the gene controls the phenotype, use of the conditional
distribution P(x|a), i.e. the distribution of phenotype for
given genotype a would be natural, to compute P(x,a) as
p(a)P(x|a). In contrast, the eq. (5) takes symmetric form
with regards to x and a. Here the form is chosen after taking
into account of selection process based on phenotype x. This
selection process introduces a feedback from phenotype to
the genotype distribution as in figure 2. Considering quasistatic evolutionary process to select phenotype x, the form
(5) is chosen for P(x,a).
Third, the stability assumption is introduced for robust
and gradual evolution, postulating that P(x,a) has a single
peak in (x,a) space at each generation. Fourth, existence
of threshold mutation rate μ max is implicitly assumed,
beyond which the stability condition is not satisfied. In
other words, existence of error catastrophe is implicitly
assumed. Such catastrophe does not exist for every coupling
Relationship among phenotypic plasticity, phenotypic fluctuations, robustness, and evolvability
form in eq. (5). For example, if we adopted the distribution
exp(−
−(x −c (a −a 0 ))2
2α
−
(a −a 0 )2
2μ
),
instead, the stability condition would always be satisfied,
as long as α > 0 and μ > 0. In other words, we assumed
existence of such loss of stability with the increase of the
mutation rate, to choose the coupling form in eq. (5). (see
also a remark below).
Fifth, to get the proportionality (10), additional
assumption is made on the smallness of the mutation
rate μ, as well as the constancy of μ and μ max through the
evolutionary course.
Note again we have used the expansion up to the second
order around a0 and X0 in eq. (5), i.e. the Gaussian-like
distribution as well as the coupling term C(x–X0)(a–a0).
In fact, this specific form is not necessary. By taking a
generalized from, the single-peaked distribution assumption
leads to the Hessian condition for logP(x,a) around the peak
at a = a0 and x = X0. This leads to the inequality (11)4. For
example, when we take the coupling C′(x–X0)(a–a0) / α in eq.
(5), the inequality (8) is replaced by μ < αC′2=μ max, whereas
the equations on Vg is replaced by Vg = μ C′2 / (1–μC′2/α), so
that the final form in eq. (9) is identical.
Remark: At μ = μ max, the distribution (5) becomes
flat. In fact, such loss of stability of a fitness distribution
roughly corresponds to error catastrophe by Eigen and
Schuster (1979), where non-fitted mutants accumulate with
the increase of mutation rate, due to their combinatorial
explosion. When fitted genotypes are rare, they are not
sustained against the increase of the mutation rate.
In our formulation, however, we cannot tell anything
about the distribution when the inequality (8) (or equivalently
eq. [11]) is not satisfied. In the formulation with Gaussian
distribution, the distribution of phenotype is symmetric around
the peak. However, near this catastrophic point, the variance
increases so that the expansion up to the second order in eq. (5)
will no longer be valid. In general, the genetic sequence giving
rise to fitted phenotype is rare, and thus the distribution at
μ ~ μ max is expected to be extended mostly to the lower fitness
side, due to possible higher order terms in the expansion of
a–a0 and x–X0. (In other words, this asymmetry to the side
of lower fitness values in the distribution is also expected
from the projection to a one-dimensional scalar variable a
from a genetic sequence.) Accordingly, it is expected that
the evolution to select a higher fitness phenotype no longer
works when the inequality is not satisfied, as discussed by
Eigen as error catastrophe.
A remark should be made here: In the discussion of
Eigen, phenotype is uniquely determined by the genotype.
In contrast, the present error catastrophe is associated
4
Note Vg in (Kaneko and Furusawa 2006) needs to be interpreted
as Vig here, as remarked in Kaneko and Furusawa (2008).
J. Biosci. 34(4), October 2009
535
with stochasticity in genotype-phenotype mapping,
as demonstrated by Vip > 0, which is a result of noise
encountered during the developmental dynamics. In fact,
in the example to be discussed in the next section, the
catastrophe occurs as a result of the decrease in the noise
strength in developmental dynamics. In this sense, the error
catastrophe here is not identical with that by Eigen.
4.
Microscopic approach: Gene regulation network
model
As the above discussion on phenotypic variances on genetic
and epigenetic origins is based on several assumptions,
it is important to study the evolution of robustness
and phenotypic fluctuations, by using a model with
“developmental” dynamics of phenotype. To introduce
such dynamical systems model, one needs to consider the
following structure of evolutionary processes.
(i) There is a population of organisms with a distribution
of genotypes.
(ii) Phenotype is determined by genotype, through
“developmental” dynamics.
(iii) The fitness for selection, i.e. the offspring number,
is given by the phenotype.
(iv) The distribution of genotypes for the next
generation changes according to the change of
genotype by mutation, and selection process
according to the fitness.
During evolution in silico, procedures (i), (iii), and (iv) are
adopted as genetic algorithms. Here, however, it is important
that the phenotype is determined only after (complex)
‘developmental’ dynamics (ii), which are stochastic due to
the noise therein.
We previously studied a protocell model with catalytic
reaction network in which cells that have a higher
concentration of a specific chemical are selected among
the networks having different paths. Extensive numerical
simulations confirm the inequality Vip > Vig, error catastrophe
at Vip ~ Vig, and the proportionality between Vip and Vg. Note
that in the numerical evolution of this model, none of the
five assumptions in the last section are adopted.
As another example, we have studied gene expression
dynamics that are governed by regulatory networks (Glass
and Kauffman 1973; Mjolsness et al. 1991; Salzar-Ciudad
et al. 2000). The developmental process (ii) is given by this
gene expression dynamics. It shapes the final gene expression
profile that determines the evolutionary fitness. Mutation
at each generation alters the network slightly. Among the
mutated networks, we select those with higher fitness values.
To be specific, the dynamics of a given gene expression
level xi is described by:
M
dx i /dt = γ( f (∑ J ij x j ) − x i ) + σηi (t ),
j
(12)
536
Kunihiko Kaneko
Figure 2. Schematic representation of genotype-phenotype relationship. The genotype a is distributed by p(a), whereas there is a
stochastic mapping from the genotype to phenotype as a result of developmental dynamics with noise. Thus the phenotype x is distributed
even for isogenic individuals, so that the distribution P(x,a) is defined. Mutation-selection process works on phenotype x, so that a feedback
from P(x,a) to the genotype distribution P(a) is generated.
where Jij = –1,1,0, and ηi(t) is Gaussian white noise given by
<ηi(t)ηj(t′)> = δi,jδ(t–t′). M is the total number of genes. The
function f represents gene expression dynamics following
Hill function, or described by a certain threshold dynamics
(Kaneko 2007, 2008).
The value of σ represents the strength of noise encountered
in gene expression dynamics, originated in molecular
fluctuations in chemical reactions. We prefix the initial
condition for this “developmental” dynamical system (say,
at the condition that none of the genes are expressed). The
phenotype is a pattern of {xi} developed after a certain time.
The fitness F is determined as a function of xi. Sometimes,
not all genes contribute to the fitness. It is given by a
function of only a set of “output” gene expressions. In the
paper (Kaneko 2007), we adopted the fitness as the number
of expressed genes among the output genes j = 1, 2,…, k <
M, i.e. xj greater than a certain threshold. Selection is applied
after the introduction of mutation at each generation in the
transcriptional regulation network, i.e. the matrix Jij.
Among the mutated networks, we select a certain fraction
of networks that has higher fitness values. Because the
network Jij determines the ‘rule’ of the dynamics, it is natural
to treat Jij as a measure of genotype. Individuals with different
genotypes have different sets of Jij.
J. Biosci. 34(4), October 2009
As the model contains a noise term, whose strength is
given by σ, the fitness can fluctuate among individuals sharing
the same genotype Jij. Hence the fitness F is distributed. This
leads to the variance of the isogenic phenotypic fluctuation
denoted by Vip ({J}) for a given genotype (network) {J}, i.e.
Vip ({J}) =
∫ dFPˆ(F ;{J})(F − F ({J})) ,
2
(13)
where P̂ (F;{J}) is the fitness distribution over isogenic
species sharing the same network Jij, and F̄ ({J}) = ∫FP̂
(F; {J})dF is the average fitness. To check the relationship
between Vip and Vg, we either use the Vip for {J} that gives
the peak value of the fitness distribution or the average V̄ip
= ∫d {J}p ({J}) Vip ({J}), where p({J}) is the population
distribution of the genotype {J}. In both cases, the same result
on the relationship between Vip and Vg is obtained. Here we
briefly describe the result with V̄ ip. At each generation, there
exists a population of individuals with different genotypes
{J}. On the other hand, the phenotypic variance by genetic
distribution is given by
Vig =
∫ d{J}p({J})(F (J )− < F >) ,
2
(14)
Relationship among phenotypic plasticity, phenotypic fluctuations, robustness, and evolvability
where <F̄> = ∫ p({J}) F̄ ({J})d{J} is the average of average
fitness over all populations. (Here, as we continuously select
fitted phenotypes, even if they are rare, the remaining
population mainly consists only of those having nearly the
optimal fitness value. Thus we expect that the variance by
the above genetic algorithm roughly estimates Vig rather than
Vg, according to the formulation of the last section.)
Results from several evolutionary simulations of this
class of models are summarized as follows:
(i) There is a certain threshold noise level σc, beyond
which the evolution of robustness progresses,
so that both Vig and V̄ ip decrease. There, most of
the individuals take the highest fitness value. In
contrast, for a lower level of noise σ < σc, mutants
that have very low fitness values always remain.
There are many individuals taking the highest
fitness value, whereas the fraction of individuals
with much lower fitness values does not decrease.
Hence the increase of the average fitness value over
the total population stops at a certain level.
(ii) At around the threshold noise level, Vig approaches
V̄ ip and at σ < σc, Vig ~ V̄ ip holds, whereas for
σ > σc, V̄ ip > Vig is satisfied. For robust evolution to
progress, this inequality is satisfied.
(iii) When noise is larger than this threshold, Vig~
Vg∝V̄ ip holds (see figure 3), through the evolution
course after a few generations. (Here as Vig << Vip,
the difference between Vig and Vg is negligible).
Thus, the results are consistent with those given by the
phenomenological theory in the last section. Here, however,
we should note that there is not a direct correspondence
with the phenomenological distribution theory and the gene
expression network model. So far no direct derivation of
the former from the latter is available: The genotype here
is given by the matrix {J}, and is not a priori represented
by a scalar parameter a. The accumulation of low-fitness
mutants with the decrease of noise level corresponds to
the error catastrophe at Vip ~ Vig but existence of such error
catastrophe itself is not a priori assumed in the model.
The proportionality of the two variances is not evident
in the model, either. Indeed, as mentioned above, the
proportionality is observed only after some generations,
when evolution progresses steadily. This might correspond
to the assumption for the quasi-static evolution of P(x,a).
Why does the system not maintain the highest fitness state
under a small phenotypic noise level with σ < σc? Indeed, the
dynamics of the top-fitness networks that evolved under
such low noise level have distinguishable dynamics from
those that evolved under a high noise level. It was found
that, for networks evolved under σ < σc, a large portion of
the initial conditions reach attractors that give the highest
fitness values, while for those evolved under σ < σc, only a
J. Biosci. 34(4), October 2009
537
tiny fraction (i.e. in the vicinity of all-off states) is attracted
to them.
When the time course of gene expression dynamics to
reach its final pattern (attractor) is represented as a motion
falling along a potential valley, our results suggest that
the potential landscape becomes smoother and simpler
through evolution and loses its ruggedness after a few dozen
generations, for σ > σc. In this case, the ‘developmental’
dynamics gives a global, smooth attraction to the target (see
figure 4). In fact, such type of developmental dynamics with
global attraction is known to be ubiquitous in protein folding
dynamics (Abe and Go 1980; Onuchic et al. 1995), gene
expression dynamics(Li et al. 2004), and signal transduction
network (Wang et al. 2006). On the other hand, the landscape
evolved at σ < σc is rugged. Except from the vicinity of given
initial conditions, the expression dynamics do not reach the
target pattern, as in the lower column of figure 4.
Now, consider mutation to a network to alter slightly a few
elements Ji, j of the matrix J. This introduces slight alterations
in gene expression dynamics. In smooth landscape with
global basin of attraction, such perturbation in dynamics
makes only little change to the final expression pattern
(attractor). In contrast, under the dynamics with rugged
developmental landscape, a slight change easily destroys the
attraction to the target attractor. Then, low-fitness mutants
appear. This explains why the network evolved at a low noise
level is not robust to mutation. In other words, evolution to
eliminate ruggedness in developmental potential is possible
only for a sufficiently high noise level.
We expect these relationships (i)–(iii) as well as the
existence of the error catastrophe to be generally valid for
systems satisfying the following conditions.
(A)
Fitness is determined through developmental
dynamics.
(B) Developmental dynamics is complex so that it may
fail to produce the target phenotype that gives a
high fitness value, when perturbed by noise.
(C) There is effective equivalence between mutation
to gene and noise in the developmental dynamics,
with regards to phenotypic change.
Condition (A) is generally satisfied for any biological
system. Even in a unicellular organism, phenotype is shaped
through gene expression dynamics. The condition (B) is
the assumption on the existence of ’error catastrophe’. A
phenotypic state with a higher function is rather rare, and it
is not easy to reach such a phenotype through the dynamics
involving many degrees of freedom. In our model, the
condition (B) is provided by the complex gene expression
dynamics to reach a target expression pattern. In a system
with (A)–(B), slight differences in genotype may lead
to a large differences in phenotype, as the phenotype is
determined after temporal integration of the developmental
Kunihiko Kaneko
538
dynamics where slight differences in genotype (i.e. in the
rule that governs the dynamics) are accumulated through
the course of development and amplified. This introduces
“error catastrophe” as a result of noise in developmental
dynamics.
Postulate (C) also seems to be satisfied generally; by noise,
developmental dynamics are modulated to change a path in
the developmental course, whereas changes introduced by
mutation into the equations governing the developmental
dynamics can have a similar effect. In the present model, the
phenotypic fluctuation to increase synthesis of a specific gene
product and mutation to add another term to increase such
production can contribute to the increase in the expression
level of a given gene in a similar manner.
5.
Discussion
Our theory has several implications in development and
evolution. We briefly discuss a few topics.
5.1
Evolvability and its relationship with fluctuations:
answer to question 1
In section 2, we have proposed the proportionality between
phenotypic plasticity and phenotypic fluctuation Vip as a
natural consequence of fluctuation-response relationship.
Then we have proposed the proportionality between this
epigenetic phenotypic fluctuation Vip and the fluctuation
due to genetic variation, Vg. Since the latter is proportional
1
0.1
to the evolution speed according to Fisher’s theorem, the
phenotypic plasticity and evolvability are correlated through
the phenotypic fluctuation. Although such relationship was
discussed qualitatively over decades, quantitative conclusion
has not been drawn yet. Here we have proposed such
quantitative relationship by phenomenological distribution
theory, to be consistent with the data from numerical and in
vivo experiments.
Waddington proposed genetic assimilation in which
phenotypic change due to the environmental change is
later embedded into genes (Waddington 1953, 1957). The
proportionality among phenotypic plasticity, Vip, and Vg
is regarded as a quantitative expression of such genetic
assimilation, since Vg gives an index for the evolution speed.
Evolvability may depend on species. Some species called
living fossil preserve their phenotype over much longer
generations than others. An origin of such difference in
evolvability could be provided by the degree of rigidness
in developmental process. If the phenotype generated
by developmental process is rigid, then the phenotypic
fluctuation as well as phenotypic plasticity against
environmental change is smaller. Then according to the
theory here, the evolution speed will be smaller.
Evolvability depends not only on species but on each gene
in a given species. Some proteins could evolve more rapidly
than others. If we apply the fluctuation-response relationship
in section 2 the expression level of each gene, correlation
between the level of fluctuation of each gene expression and
its evolution speed is expected to exist not only through the
evolutionary course, but also for different genes in a given
σ =.1
σ=.06
σ=.01
Vg=Vip
200
180
160
140
120
100
80
60
40
20
0
Vg
0.01
0.001
1e-04
1e-05
1e-06
1e-05
1e-04
0.001
0.01
0.1
1
Vip
Figure 3. The relationship between Vg (Vig) and V̄ ip. Vg is computed from the distribution of fitness over different {J} at each generation
(eq. [14] , and V̄ ip by averaging the variance of isogenic phenotype fluctuations over all existing individuals (eq. [13]). Computed by using
the model eq. [12] (see [Kaneko 2007] for detailed setup). Points are plotted over 200 generations, with gradual colour change as displayed.
σ = 0.01 (○), 0.06 (×), and 0.1 (■). For σ > σc ≈ 0.02, both decrease with successive generations.
J. Biosci. 34(4), October 2009
Relationship among phenotypic plasticity, phenotypic fluctuations, robustness, and evolvability
539
Figure 4. Schematic representation of the basin structure, represented as a process of descending a potential landscape. In general,
developmental dynamics can have many attractors, and depending on initial conditions, they may reach different phenotypes, so that the
developmental landscape is rather rugged. After evolution, global and smooth attraction to the target phenotype is shaped by developmental
dynamics, if gene expression dynamics are sufficiently noisy.
generation. To discuss such dependence on each protein, we
have studied the model discussed in section 4, consisting
of expression levels of a variety of genes, and found the
proportionality between Vg and Vip over expressions of a
variety of genes. Indeed, we have measured the variances
for expression levels of each gene i, and defined their genetic
and epigenetic variances Vg (i) and Vip (i) for each. From
several simulations, we have confirmed the proportionality
between Vg (i) and Vip (i) over many genes i (Kaneko 2008).
Although direct experimental support is not available yet,
recent data on isogenic phenotypic variance and mutational
variance over proteins in yeast may suggest the existence
of such correlation between the two (Landry et al. 2007;
Lehner 2008).
5.2
Evolution of robustness and relevance of noise:
Answer to the question 2
As mentioned in the introduction, biological robustness
is the insensitivity of fitness or phenotype to system’s
changes. Thus, the developmental robustness to noise is
characterized by the inverse of Vip, while the mutational
robustness is characterized by the inverse of Vg. In Section 4,
it is shown that the two variances decrease in proportion
J. Biosci. 34(4), October 2009
through the evolution under a fixed fitness condition, if there
exits a certain level of noise in the development. Robustness
evolves, as discussed by Schmalhausen’s stabilizing
selection (Schmalhausen 1949) and Waddington’s
canalization (Waddington 1942, 1957). On the other hand,
the proportionality between the two variances implies
the correlation between developmental and mutational
robustness. Note that this evolution of robustness is possible
only under a sufficient level of noise in development.
Robustness to noise in development brings about robustness
to mutation.
Our result demonstrates the role of noise during
phenotypic expression dynamics to evolution of robustness.
Despite recent quantitative observations on phenotypic
fluctuation, noise is often thought to be an obstacle in
tuning a system to achieve and maintain a state with a
higher function, because the phenotype may be deviated
from a state with a higher function, perturbed by such
noise. Indeed, questions most often asked are how a given
biological function can be maintained in spite of the
existence of phenotypic noise (Ueda et al. 2001; Kaern et al.
2005). In contrast, we have revealed a positive role of noise
to evolution quantitatively. We also note that relevance of
noise to adaptation (Kashiwagi et al. 2006; Furusawa and
Kunihiko Kaneko
540
Kaneko 2008) and to cell differentiation (Kaneko and Yomo
1999) is recently discussed.
In our theory, robustness in genotype and phenotype are
represented by the single-peaked-ness in the distribution P(x
= phenotype, a = genotype). If we write the distribution in a
“potential form” as
P(x,a) = exp (–U(x,a)),
(15)
the stability condition means existence of global minimum
of the potential U(x,a). With the increase in the slope of the
attraction in the potential, the robustness increases. The
smooth attraction in the developmental potential shown in
Section 4 is a representation of canalization in the epigenetic
landscape by Waddington, whereas the correlation in the
attractions in phenotypic (x) and genetic (a) directions is one
representation of genetic assimilation.
5.3
Future issues
The proportionality between the phenotypic variances of
isogenic individuals and that due to genetic variation in
Section 2 and Section 3 is not a result of derivation from the
first principle, but is a result of phenomenological description
based on several assumptions. Although numerical results on
gene expression network and catalytic reaction network
models support the relationship, there remains a large gap
between the macroscopic phenomenological theory and
microscopic simulations. It should be important to establish a
statistical physics theory to fill the gap between the two (see
also Sakata et al. [2009] for a study towards this direction).
All of the plasticity, fluctuation, and evolvability decrease
in our model simulations through the course of evolution,
under selection by a fixed fitness condition. Considering
that none of the phenotypic plasticity, fluctuation, and
evolvability vanishes in the present organisms, some process
for restoration or maintenance of plasticity or fluctuation
should also exist. To consider such process, the followings
are necessary:
(1)
(2)
Fitness with several conditions: In our model study
(and also in experiments with artificial selection), the
fitness condition is imposed by a single condition. In
nature, the fitness may consist in several conditions.
In this case, we may need more variables in the
characterization of the distribution function P, say
P(x,y,a,b) etc. In such multi-dimensional theory,
interference among different phenotypic variable x
and y may contribute to the increase or maintenance
of fluctuations.
Environmental variation: If the environmental
condition changes in time, the fluctuation level
may be sustained. To mimic the environmental
change, we have altered the condition of target
J. Biosci. 34(4), October 2009
gene expression pattern in the model of Section
4 after some generation. To be specific, some of
the target genes should be “off” instead of “on”.
After this alteration in the fitness condition,
both the fluctuations Vip and Vg increase over
some generations, and then decrease to fix the
expression of target genes. It is interesting that
both the expression variances increase keeping
proportionality. At any rate, the restoration of
fluctuation emerges as a result of the change in
environmental condition. After this change, both the
variances increase to adapt the new environmental
condition (K Kaneko, unpublished). Hence, the
variances again work as a measure of plasticity. It
will also be important to incorporate environmental
fluctuation into our theoretical formulation.
(3) Interaction: Another source to sustain the level of
fluctuation is interaction among individuals, which
may introduce a change in the ‘environmental
condition’ that each individual faces with. Here as
the variation in phenotypes is larger, the variation in
the interaction will be increased. Due to the variation
in the interaction term, the phenotypic variances will
be increased. Then, mutual reinforcement between
diversity in interaction and phenotypic plasticity
may be expected, which will result in both the
genetic and phenotypic diversity. Co-evolution for
such reinforcement between phenotypic plasticity
and genetic diversity will be an important issue to
be studied in future.
(4) Speciation: So far, we have considered gradual
evolution keeping a single peak in phenotype
distribution. On the other hand, the bimodal
distribution should appear at the speciation.
Previously we have shown that robust sympatric
speciation can occur as a mutual reinforcement
of bifurcation of phenotypes as a result of
developmental dynamics and due to genetic
variation (Kaneko and Yomo 2000; Kaneko 2002).
It will be important to reformulate the problem in
terms of the present distribution theory.
(5) Sexual reproduction: Our simulation in the present
paper was based on asexual haploid population. On
the other hand, the phenomenological theory on the
distribution could be applicable also to a population
with sexual reproduction. In the sexual reproduction,
however, recombination is a major source of genetic
variation, rather than the mutation. Then, instead of
the mutation rate, the recombination rate could
control the error catastrophe. Here, the degree of
genetic heterogeneity induced by recombination
strongly depends on the genotype distribution at the
moment. (For example, for an isogenic population,
Relationship among phenotypic plasticity, phenotypic fluctuations, robustness, and evolvability
the mutation can introduce heterogeneity, but
the recombination only cannot produce the
heterogeneity.) It would be interesting to extend
our theory on evolutionary stability and phenotypic
fluctuation to incorporate the sexual reproduction.
6.
Conclusion
We have proposed general relationship among phenotypic
plasticity and phenotypic fluctuations of genetic and
epigenetic origins. Through experiments and numerical
simulations, the relationship is confirmed. Increase
of developmental and mutational robustness through
evolution is demonstrated, which progresses keeping the
proportionality between the two.
Recall Waddington’s canalization, epigenetic landscape,
and genetic assimilation (Waddington 1957). Our distribution
function P(phenotype = x, genotype = a) = exp (–U(x,a))
and its stability give one representation of the epigenetic
landscape, and canalization. The decrease in fluctuations
through evolution means that the potential valley is deeper,
as described by canalization. The genetic assimilation is
represented by the proportionality between the fluctuations
of genetic and epigenetic origins. Note that the existence of
the phenotype-genotype distribution P(x,a) = exp(–U(x,a))
is based on mutual relationship between phenotype and
genotype, The genotype determines (probabilistically) the
phenotype through developmental dynamics, while the
phenotype restricts the genotype through mutation-selection
process. For a robust evolutionary process to progress, the
consistency between phenotype and genotype is shaped,
which is the origin of the distribution P (phenotype, genotype)
assumptions, and the general relationships proposed here.
In formulating fluctuation-response relationship by
Einstein, such “consistency” between microscopic and
macroscopic descriptions was a guiding principle. We hope
that quantitative formulation of Waddington’s ideas based
on Einstein’s spirit will provide a coherent understanding in
evolution-development relationship.
Acknowledgements
I would like to thank C Furusawa, T Yomo, K Sato, M
Tachikawa, S Ishihara, S Sawai for continual discussions, an
anonymous referee for important comments and E Jablonka,
T Sato, S Jain, S Newman, and V Nanjundiah for stimulating
discussions during the conference at Trivandrum.
The theoretical study here was inspired at 2005,
a hundredth anniversary both of Einstein’s miracle year and
of Waddington’s birth, advanced through the conference
held in 2007, the 50th anniversary of the publication of the
Strategy of the Genes by Waddington, and the manuscript
is completed at the 200th anniversary of Darwin’s birth.
J. Biosci. 34(4), October 2009
541
I would like to express sincere gratitude to these giants for
inspiration.
References
Abe H and Go N 1980 Noninteracting local-structure model
of folding and unfolding transition in globular proteins. I.
Formulation; Biopolymers 20 1013
Alon U, Surette M G, Barkai N and Leibler S 1999 Robustness in
bacterial chemotaxis; Nature (London) 397 168–171
Ancel L W and Fontana W 2002 Plasticity, evolvability, and
modularity in RNA; J. Exp. Zool. 288 242–283
Bar-Even A, Paulsson J, Maheshri N, Carmi M, O’Shea E, Pilpel
Y and Barkai N 2006, Noise in protein expression scales with
natural protein abundance; Nat. Genet. 38 636–643
Barkai N and Leibler S 1997 Robustness in simple biochemical
networks; Nature (London) 387 913–917
Callahan H S, Pigliucci M and Schlichting C D 1997 Developmental
phenotypic plasticity: where ecology and evolution meet
molecular biology; Bioessays 19 519–525
Ciliberti S, Martin O C and Wagner A 2007 Robustness can evolve
gradually in complex regulatory gene networks with varying
topology; PLoS Comp. Biol. 3 e15
de Visser J A, Hermisson J, Wagner G P, Ancelmeyers L, BagheriChichian H, Blanchard J L, Lin C, Cheverud J M. et al. 2003
Evolution and detection of genetic robustness; Evolution 57
1959–1972
Eigen M and Schuster P 1979 The hypercycle (Heidelberg:
Springer)
Einstein A 1926 Investigation on the Theory of of Brownian
Movement (Collection of papers ed. by R Furth), Dover
(reprinted 1956)
Elowitz M B, Levine A J, Siggia E D and Swain P S 2002
Stochastic gene expression in a single cell; Science 297 1183–
1187
Fisher R A 1930 The genetical theory of natural selection (Oxford
University Press) (reprinted 1958)
Furusawa C and Kaneko K 2008 A generic mechanism for adaptive
growth rate regulation; PLoS Comput. Biol 4 e3
Furusawa C, Suzuki T, Kashiwagi A, Yomo T and Kaneko K 2005
Ubiquity of log-normal distributions in intra-cellular reaction
dynamics; Biophysics 1 25–31
Gibson G and Wagner G P 2000 Canalization in evolutionary
genetics: a stabilizing theory?; Bioessays 22 372–380
Glass L and Kauffman S A 1973 The logical analysis of continuous,
non-linear biochemical control networks; J. Theor. Biol. 39
103–129
Hasty J, Pradines J, Dolnik M and Collins J J 2000 Noise-based
switches and amplifiers for gene expression; Proc. Natl. Acad.
Sci. USA 97 2075–2080
ItoY, Toyota H, Kaneko K and Yomo T 2009 How evolution affects
phenotypic fluctuation; Mol. Syst. Biol. 5 264
Kaern M, Elston T C, Blake W J and Collins J J 2005 Stochasticity
in gene expression: from theories to phenotypes; Nat. Rev.
Genet. 6 451–464
Kaneko K 2006 Life: An introduction to complex systems biology
(Heidelberg and New York: Springer)
542
Kunihiko Kaneko
Kaneko K 2007 Evolution of robustness to noise and mutation in
gene expression dynamics; PLoS ONE 2 e434
Kaneko K 2008 Shaping Robust system through evolution; Chaos
18 026112
Kaneko K and Furusawa C 2006 An evolutionary relationship
between genetic variation and phenotypic fluctuation; J. Theor.
Biol. 240 78–86
Kaneko K and Furusawa C 2008 Consistency principle in biological
dynamical systems; Theor. Biosci. 127 195–204
Kaneko K and Yomo T 1999 Isologous diversification for robust
development of cell society; J. Theor. Biol.199 243–256
Kaneko K and Yomo T 2000 Sympatric Speciation: compliance
with phenotype diversification from a single genotype; Proc. R.
Soc. B London 267 2367–2373
Kaneko K 2002 Symbiotic sympatric speciation: Compliance
with interaction-driven phenotype differentiation from a single
genotype; Popul. Ecol. 44 71–85
Kashiwagi A, Urabe I, Kaneko K and Yomo T 2006 Adaptive
response of a gene network to environmental changes by
attractor selection; PLoS ONE 1 e49
Kirschner M W and Gerhart J C 2005 The plausibility of life (Yale
University Press)
Krishna S, Banerjee B, Ramakrishnan T V and Shivashankar G V
2005 Stochastic simulations of the origins and implications of
long-tailed distributions in gene expression; Proc. Natl. Acad.
Sci. USA 102 4771–4776
Kubo R, Toda M and Hashitsume N 1985 Statistical Physics II:
(English translation; Springer).
Landry C R, Lemos B, Rifkin S A, Dickinson W J and Hartl D L
2007 Genetic properties influencing the evolvability of gene
expression; Science 317 118–121
Lehner B 2008 Selection to minimize noise in living systems and
its implications for the evolution of gene expression; Mol. Syst.
Biol. 4 170
Li F, Long T, Lu Y, Ouyang Q and Tang C 2004 The yeast cellcycle network is robustly designed; Proc. Natl. Acad. Sci. USA
101 10040–10046
McAdams H H and Arkin A 1997 Stochastic mechanisms in gene
expression; Proc. Natl. Acad. Sci. USA 94 814–819
Mjolsness E, Sharp D H and Reinitz J 1991 A connectionist model
of development; J. Theor. Biol. 152 429–453
Onuchic J N, Wolynes P G, Luthey-Schulten Z and Socci
N D 1995 Toward an outline of the topography of a realistic
protein-folding funnel; Proc. Natl. Acad. Sci. USA 92
3626–3630
Oosawa F 1975 Effect of the field fluctuation on a macromolecular
system; J. Theor. Biol. 52 175
Pigliucci M, Murren C J and Schlichting C D 2006 Phenotypic
plasticity and evolution by genetic assimilation; J. Exp. Biol.
209 2362–2367
Salazar-Ciudad I, Garcia-Fernandez J and Sole R V 2000 Gene
networks capable of pattern formation: from induction to
reaction–diffusion; J. Theor. Biol. 205 587–603
Sakata A, Hukushima K and Kaneko K 2009 Funnel landscape
and mutational robustness as a result of evolution under thermal
noise; Phys. Rev. Lett. 102 148101
Sato K and Kaneko K 2007 Evolution equation of phenotype
distribution: general formulation and application to error
catastrophe; Phys. Rev. E 75 061909
Sato K, Ito Y, Yomo T and Kaneko K 2003 On the relation between
fluctuation and response in biological systems; Proc. Natl.
Acad. Sci. USA 100 14086–14090
Schmalhausen I I 1949 Factors of evolution: The theory of
stabilizing selection (Chicago: University of Chicago Press)
(reprinted 1986).
Spudich J L and Koshland D E Jr 1976 Non-genetic individuality:
chance in the single cell; Nature (London) 262 467–471
Ueda M, Sako Y, Tanaka T, Devreotes P and Yanagida T
2001 Single-molecule analysis of chemotactic signaling in
Dictyostelium cells; Science 294 864–867
Waddington C H 1942 Canalization of development and the
inheritance of acquired characters; Nature (London) 150 563–565
Waddington C H 1953 Genetic assimilation of an acquired
character; Evolution 7 118–126
Waddington C H 1957 The strategy of the genes (London:Allen
and Unwin)
Wagner A 2000 Robustness against mutations in genetic networks
of yeast; Nat. Genet. 24 355–361
Wagner A 2005 Robustness and evolvability in living systems
(Princeton, NJ: Princeton University Press)
Wagner G P, Booth G and Bagheri-Chaichian H 1997 A population
genetic theory of canalization; Evolution 51 329–347
Wang J, Huang B, Xia X and Sun Z 2006 Funneled landscape leads
to robustness of cellular networks: MAPK signal transduction;
Biophys. J. 91 L54–L56
Weinig C 2000 Plasticity versus canalization: Population
differences in the timing of shade-avoidance responses;
Evolution 54 441–451
West-Eberhard M J 2003 Developmental plasticity and evolution
(Oxford: Oxford University Press)
ePublication: 4 September 2009
J. Biosci. 34(4), October 2009